Monitoring Osteoporosis Therapy with Bone Densitometry: A Vital Tool or Regression toward Mediocrity?

In a recent issue of the Journal of the American Medical Association, Cummings et al. (1) demonstrated the phenomenon of regression to the mean when osteoporosis therapy was monitored by bone densitometry using data from two major clinical trials, the Fracture Intervention Trial (FIT) and the Multiple Outcomes of Raloxifene Evaluation (MORE) trial. Bone density data from the first 2 yr of each trial in compliant women only was reviewed. In FIT, 2634 women in the intervention group and 2603 women in the control group had the requisite bone density data and pill count to be included in the analysis. In the MORE trial, 3954 women met the criteria for inclusion in this analysis. Briefly, the authors reported that 82% of the compliant women in the intervention group from FIT had an apparent gain in total hip bone mineral density (BMD), whereas only 1.4% had an apparent loss of more than 4% in the first year. In the MORE trial, 75% of the compliant women in the intervention group had an apparent gain in femoral neck BMD, whereas 8% had an apparent loss of more than 4% in the first year. In both cases, those women who lost the most during the first year were observed to be the most likely to gain BMD during the second year. Conversely, those women with the greatest gains in BMD during the first year were most likely to lose BMD or have little, if any, gain during the second year. The authors stated that similar findings were noted at the spine in the intervention group from FIT and at both skeletal sites in the placebo group in FIT. The authors attributed the gains in bone density in the overwhelming majority of women in the intervention groups of both studies to the efficacy of the drugs used. The finding of gains in bone density in the second year even after an apparent decline in the first year was additional proof of efficacy, according to the authors. That great gains in the first year were followed by losses in the second year and great losses in the first year were followed by gains in the second year was attributed to a phenomenon known as “regression to the mean.” The authors reasoned that monitoring changes in bone density in women on alendronate or raloxifene was of questionable value because virtually all compliant women on these agents gain bone density eventually. They noted that it was likely that the phenomenon of regression to the mean was responsible for the paradoxical swings in bone density in compliant women, rather than drug failure or drug resistance. Regression to the mean, they theorized, likely exists for other quantitative measurements such as lipids, blood pressure, and spirometry values, as well as bone density. Monitoring these and other types of parameters of treatment in clinical practice are likely affected by regression to the mean. Regression to the mean was first described by Sir Francis Galton in 1886 (2). In a study of the hereditary nature of height using a population of parents and their children, Galton observed that the children of the taller parents had a mean height that was closer to the mean height of all children than the mean height of their parents was to the mean height of all parents. Similarly, if one looked at the taller children, the mean height of their parents was closer to the mean height of all parents than was the mean height of these children to the mean of all the children. Galton concluded, “Each peculiarity in a man is shared by his kinsman, but on the average in a less degree.” Galton called this “regression towards mediocrity.” Today, it is called regression to the mean. Regression to the mean will occur whenever a variable is measured on two separate occasions, when the value of the variable can change in the individual either due to biologic or measurement variability or both, and when a subgroup of the whole group is defined on the basis of a high or low value at the first measurement (3). In these circumstances, the average value for the second measurement of the variable in the subgroup will always be closer to the mean than was the first. In the FIT and MORE trials individuals with low bone density were selected (4, 5). A variable bone density was measured. The bone density in an individual could change, either because of the imperfect precision of bone density measurements, the effect of the therapeutic agent, or both. In the analysis by Cummings et al. (1), the individuals were then divided into subgroups based on the apparent change in bone density during the first year of each study. These conditions met the necessary requirements to create the phenomenon of regression to the mean when the change in bone density in the second year was determined for each subgroup. In these circumstances, regression to the mean occurred with bone density measurements, just as Galton could have predicted it would, so long ago. The change in bone density during the second year of each study regressed toward the mean change for the entire group at the end of the Received August 14, 2000. Accepted August 14, 2000. Address correspondence and requests for reprints to: Sydney Lou Bonnick, M.D., FACP, Institute for Women’s Health, Texas Woman’s University, P.O. Box 425876, Denton, Texas 76204. 0021-972X/00/$03.00/0 Vol. 85, No. 10 The Journal of Clinical Endocrinology & Metabolism Printed in U.S.A. Copyright © 2000 by The Endocrine Society